A software server configuration method and system for user usage feature analysis

Abstract

The application provides a software server configuration method and system for user feature analysis, and belongs to the technical field of servers. The method comprises the following steps: constructing a hierarchical distributed content service architecture; monitoring and obtaining a user behavior data set, identifying and determining a user access feature, updating second hot data according to the user access feature, and storing the updated second hot data to a cloud server cache layer; setting an adaptive dimension reduction strategy according to the user access feature, reducing the dimension of the updated second hot data, obtaining updated first hot data, and storing the updated first hot data to an edge processing layer; continuously monitoring the cache hit state of a user access request, obtaining a first data miss rate and a second data miss rate in a preset time zone, and triggering a cache content adjustment mechanism or optimizing and adjusting the updated second hot data and / or the updated first hot data. The multi-layer cache dynamic optimization is realized through user behavior perception and intelligent dimension reduction strategy, and the cache hit rate and response speed are improved.

Description

Technical Field

[0001] This invention relates to the field of server technology, and in particular to a software server configuration method and system for user usage characteristic analysis. Background Technology

[0002] With the rapid development of internet technology and the continuous growth of user scale, content service systems face the challenge of massive user requests and high concurrency access. To improve the response speed of such systems and reduce database load, existing technologies widely adopt caching techniques, which accelerate data access by pre-storing frequently accessed data in memory.

[0003] Currently, common caching architectures mainly include single-layer caching and multi-layer caching. Single-layer caching typically uses in-memory databases such as Redis to store frequently accessed data in memory; multi-layer caching adds distributed caching layers such as CDN edge nodes on top of single-layer caching, forming a hierarchical data storage structure.

[0004] However, existing caching configurations mostly employ static preset strategies, failing to dynamically adjust based on actual user access behavior and characteristics. This static configuration approach struggles to adapt to changes in user access patterns, leading to a mismatch between cached data and actual user needs. Secondly, in multi-layered caching architectures, the lack of effective coordination mechanisms between different caching layers prevents differentiated content configuration based on storage characteristics and user access patterns, resulting in wasted cache resources and reduced hit rates. Furthermore, existing technologies lack real-time caching performance monitoring and feedback mechanisms, failing to promptly detect declining cache hit rates or dynamically optimize caching configuration strategies based on hit rate changes.

[0005] The aforementioned technical deficiencies result in a low hit rate for the existing cache, causing frequent user requests to return to the database, leading to increased response latency and excessive database load, which seriously affects the overall service quality and user experience. Summary of the Invention

[0006] This invention addresses the technical problem in existing technologies that cannot dynamically adjust the configuration of multi-layer cache content based on user access characteristics, resulting in low cache hit rate and high response latency. It provides a software server configuration method and system based on user usage characteristic analysis to solve this problem.

[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0008] In a first aspect, the present invention provides a software server configuration method for user usage feature analysis, comprising: constructing a layered distributed content service architecture, wherein the layered distributed content service architecture includes an edge processing layer, a cloud server cache layer, and a data center persistence layer, the edge processing layer is used to store first hot data, and the cloud server cache layer is used to store second hot data; monitoring and acquiring user behavior datasets within a historical time window, identifying and determining user access features, updating the second hot data according to the user access features, and storing the updated second hot data in the cloud server cache layer; setting an adaptation dimensionality reduction strategy according to the user access features, reducing the dimensionality of the updated second hot data, and storing the updated first hot data in the edge processing layer; continuously monitoring the cache hit status of user access requests, acquiring the first data miss rate and the second data miss rate within a preset time zone, and if the first data miss rate and / or the second data miss rate are greater than a preset threshold, triggering a cache content adjustment mechanism, and optimizing and adjusting the updated second hot data and / or the updated first hot data based on reinforcement learning.

[0009] Secondly, the present invention provides a software server configuration system for user usage feature analysis, comprising: a layered architecture construction module for constructing a layered distributed content service architecture, wherein the layered distributed content service architecture includes an edge processing layer, a cloud server cache layer, and a data center persistence layer, the edge processing layer for storing first hot data, and the cloud server cache layer for storing second hot data; a user behavior monitoring module for monitoring and acquiring user behavior datasets within a historical time window, identifying and determining user access features, updating the second hot data according to the user access features, and storing the updated second hot data in the cloud server cache layer; a data dimensionality reduction processing module for setting an adaptation dimensionality reduction strategy according to the user access features, reducing the dimensionality of the updated second hot data, and storing the updated first hot data in the edge processing layer; and a cache optimization and adjustment module for continuously monitoring the cache hit status of user access requests, acquiring a first data miss rate and a second data miss rate within a preset time zone, and if the first data miss rate and / or the second data miss rate are greater than a preset threshold, triggering a cache content adjustment mechanism to optimize and adjust the updated second hot data and / or the updated first hot data based on reinforcement learning.

[0010] The beneficial effects of this invention are:

[0011] A layered distributed content service architecture is constructed, comprising an edge processing layer, a cloud server caching layer, and a data center persistence layer. The edge processing layer stores the most frequently accessed data, while the cloud server caching layer stores the next most frequently accessed data. This layered architecture allows for tiered storage based on data access frequency and importance, laying the foundation for subsequent dynamic configuration. User behavior datasets within historical time windows are monitored and acquired. Data analysis identifies user access characteristics, and the second most frequently accessed data is updated based on these characteristics. This updated second most frequently accessed data is then stored in the cloud server caching layer, enabling intelligent cached content configuration based on actual user access behavior, making cached content more closely aligned with user needs. An adaptive dimensionality reduction strategy is set based on user access characteristics to reduce the dimensionality of the updated second most frequently accessed data, obtaining the updated most frequently accessed data, and storing it in the edge processing layer. By applying this dimensionality reduction strategy, the most critical hot data can be stored at the edge layer, enabling differentiated configuration of different caching levels and improving edge cache hit efficiency. The system continuously monitors the cache hit status of user access requests, obtains the first data miss rate and the second data miss rate within a preset time zone, and triggers a cache content adjustment mechanism when the first data miss rate and / or the second data miss rate exceed a preset threshold. Based on the reinforcement learning algorithm, the system optimizes and adjusts the updated second hot data and / or the updated first hot data, ensuring that the cache configuration can be continuously optimized according to the actual effect and dynamically improve the overall system performance.

[0012] The above technical solution enables dynamic configuration and intelligent optimization of multi-layered cache content based on user access characteristics, thereby achieving the technical effects of improving cache hit rate, reducing system response latency, and enhancing user access experience. Attached Figure Description

[0013] Figure 1 A flowchart illustrating a software server configuration method for user feature analysis provided by the present invention;

[0014] Figure 2 This is a schematic diagram of the structure of a software server configuration system for user feature analysis provided by the present invention.

[0015] In the attached diagram, the components represented by each number are as follows:

[0016] Layered architecture construction module 11, user behavior monitoring module 12, data dimensionality reduction processing module 13, and cache optimization and adjustment module 14. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0020] Example 1, as Figure 1 As shown, this embodiment of the invention provides a software server configuration method for user feature analysis, including:

[0021] S1. Construct a layered distributed content service architecture, wherein the layered distributed content service architecture includes an edge processing layer, a cloud server caching layer, and a data center persistence layer, wherein the edge processing layer is used to store first hot data, and the cloud server caching layer is used to store second hot data.

[0022] Specifically, a layered distributed content service architecture is constructed, which adopts a three-layer distributed deployment approach, consisting of an edge processing layer, a cloud server caching layer, and a data center persistence layer from top to bottom.

[0023] The edge processing layer is deployed at network edge nodes close to the user, such as CDN nodes or servers near the user. Its main functions are to quickly respond to user requests, perform initial caching, and store content. This layer stores the most frequently accessed data, i.e., high-frequency access data that has undergone dimensionality reduction processing, characterized by relatively small data volume and extremely fast access speed. For example, the edge processing layer can use a Redis cluster for data storage, combined with Nginx as a load balancer and request distribution component, to provide users with millisecond-level data access response through proximity deployment.

[0024] The cloud server caching layer is deployed in a public or private cloud environment. As an intermediate caching layer, its main function is to store frequently accessed data and handle high-concurrency access requests. This layer stores second-hot data, which is complete data of frequently accessed content identified based on user access characteristics. Compared to the first-hot data, it has more comprehensive content information. For example, the cloud server caching layer can build a distributed caching architecture based on Redis, Memcached, or other key-value caching services, and achieve high availability and horizontal scalability through clustered deployment.

[0025] The data center persistent layer is deployed in the data center as the bottom-level data storage layer, responsible for storing all complete content data, including core business data, historical data, and cold data. This layer employs diverse storage technologies; for example, structured data is managed using database systems such as MySQL and MongoDB; large-scale files and media content are stored using the HDFS distributed file system or object storage services to ensure data integrity, consistency, and persistence.

[0026] By deploying the layered distributed content service architecture, hierarchical storage and proximity access of data are achieved, ensuring both low latency for user access and data integrity and scalability.

[0027] S2. Monitor and acquire user behavior datasets within a historical time window, identify and determine user access characteristics, update the second hot data according to the user access characteristics, and store the updated second hot data in the cloud server cache layer.

[0028] Specifically, the system monitors and acquires user behavior datasets within historical time windows, identifies and determines user access characteristics, and dynamically updates the second hot data based on these characteristics.

[0029] First, within historical time windows (such as the past 7 days, 30 days, etc.), various user access behavior information is continuously collected and recorded to obtain a user behavior dataset. The user behavior dataset includes at least key indicators such as accessed content, browsing time, and access frequency. This data comprehensively reflects users' access preferences and behavioral patterns. For example, through log analysis and event tracking data collection, user access records, dwell time, and number of visits to different content pages are obtained in real time to form a user behavior dataset.

[0030] Subsequently, by analyzing and processing the collected user behavior dataset, feature information that can represent user access patterns is extracted, and user access characteristics are identified and determined. Specifically, the user behavior dataset is classified and divided according to the accessed content, and the popularity value of each content is calculated based on browsing time and access frequency. The content popularity sequence is generated by sorting the content by popularity and serves as the user access characteristics.

[0031] Next, the second hot data is updated based on user access characteristics. Specifically, based on the identified user access characteristics, the hot data content stored in the cloud server cache layer is dynamically adjusted. The content popularity sequence is comprehensively analyzed with historical popularity data, and a comprehensive content popularity sequence is calculated through gradual weighting. The top-preset proportion of high-popularity content is selected as frequently accessed content, and the complete data corresponding to these frequently accessed contents is retrieved from the data center persistence layer to form the updated second hot data.

[0032] The updated second hot data will be stored in the cloud server cache layer. This means that the new hot data, which is determined after user access feature analysis and content popularity calculation, will replace or supplement the original cached content in the cloud server cache layer. This ensures that the cache layer always stores the hot data that best meets the current user access needs, thereby improving cache hit rate and response efficiency.

[0033] S3. Based on the user access characteristics, set an adaptation dimensionality reduction strategy, perform dimensionality reduction on the updated second hot data, and obtain the updated first hot data to be stored in the edge processing layer.

[0034] Specifically, firstly, based on the user access characteristics analysis of data distribution features, an adaptation dimensionality reduction strategy is set. Specifically, the standard deviation of the content popularity sequence is calculated as the magnitude of the popularity distribution deviation, used to determine the concentration of user access behavior. When the magnitude of the popularity distribution deviation is less than or equal to a preset deviation index, it indicates that user access is relatively evenly distributed. In this case, the adaptation dimensionality reduction strategy is set to a proportional dimensionality reduction scheme, using a fixed 60% dimensionality reduction ratio to uniformly reduce the dimensionality of all frequently accessed content. When the magnitude of the popularity distribution deviation is greater than the preset deviation index, it indicates that user access exhibits a significant concentration of hotspots. In this case, the adaptation dimensionality reduction strategy is set to an adaptive dimensionality reduction scheme, assigning different adaptation dimensionality reduction ratios to different frequently accessed content. This ratio is determined based on the ratio of the average content popularity to the specific content popularity, ensuring that the higher the popularity of the content, the smaller the dimensionality reduction ratio, thus retaining more key information.

[0035] Subsequently, based on the set adaptation and dimensionality reduction strategy, the updated second-hot data is subjected to dimensionality reduction processing. Specifically, for the proportional dimensionality reduction scheme, all second-hot data content is compressed and simplified at a fixed ratio of 60%, removing redundant information but retaining core access data. For the adaptive dimensionality reduction scheme, personalized data dimensionality reduction processing is performed according to the adaptation and dimensionality reduction ratio corresponding to each high-frequency access content. Content with higher accessibility retains more detailed information, while content with relatively lower accessibility is simplified to a greater extent.

[0036] Next, dimensionality reduction is used to obtain the updated first hot data. The dimensionality-reduced data is smaller and faster to access, meeting the edge processing layer's need for extremely fast response, while preserving the core content and key information of the original second hot data. The updated first hot data is then stored at the edge processing layer. The streamlined data, processed with the adapted dimensionality reduction strategy, is deployed to various edge nodes, enabling users to access the most relevant hot content within the shortest network distance, achieving intelligent hierarchical data storage and high-speed access based on proximity.

[0037] S4. Continuously monitor the cache hit status of user access requests, obtain the first data miss rate and the second data miss rate within a preset time zone, and if the first data miss rate and / or the second data miss rate are greater than a preset threshold, trigger the cache content adjustment mechanism, and optimize and adjust the updated second hot data and / or the updated first hot data based on reinforcement learning.

[0038] Specifically, firstly, the cache hit status of user access requests is continuously monitored, and cache performance is tracked in real time. Specifically, when a user initiates a data access request, the response follows the access order of the layered distributed content service architecture: first, the first hot data in the edge processing layer is queried; if the cache hits, the data is returned directly; if the cache misses, the second hot data is queried from the cloud server cache layer; if still no cache hit, the complete data is finally obtained from the data center persistence layer. By recording the hit or miss status of each access request at each layer, complete cache hit status monitoring data is formed.

[0039] Subsequently, the first and second data miss rates within a preset time zone are obtained to quantitatively evaluate the effectiveness of each cache layer. Specifically, within a preset time zone (e.g., the past 1 hour, 6 hours, etc.), the total number of access requests and misses for the first hot data in the edge processing layer are counted to calculate the first data miss rate; similarly, relevant indicators for the second hot data in the cloud server cache layer are counted to calculate the second data miss rate. These two miss rate indicators directly reflect the degree of matching between the current cache configuration and the actual access needs of users. If the first data miss rate and / or the second data miss rate are greater than a preset threshold, the cache content adjustment mechanism is triggered. Specifically, the preset miss rate threshold (e.g., 5%, 10%, etc.) serves as the evaluation standard for cache performance. When the miss rate of any one or more layers exceeds this preset threshold, it indicates that the current cache content can no longer effectively meet the user's access needs and needs to be adjusted and optimized. At this time, the cache content adjustment mechanism is automatically triggered, and the subsequent optimization process is initiated.

[0040] Next, reinforcement learning is used to optimize and adjust the updated second-hot data and / or the updated first-hot data. Specifically, a first cache strategy optimizer and a second cache strategy optimizer are constructed, and the relationship between the miss rate and the current cache strategy is analyzed using reinforcement learning algorithms. When the first data miss rate is too high, the first cache strategy optimizer adjusts the adaptation dimensionality reduction strategy based on the miss situation, outputting an optimized dimensionality reduction strategy to improve the content selection and dimensionality reduction ratio of the first-hot data; when the second data miss rate is too high, the second cache strategy optimizer optimizes the adaptation gradient weights, adjusting the selection strategy for the second-hot data. Through the feedback mechanism of reinforcement learning, it can continuously learn and adapt to changes in user access patterns, achieving intelligent and dynamic optimization of cached content.

[0041] The above technical solution achieves intelligent optimization configuration of multi-layered caching based on user behavior analysis. This effectively solves the technical problem of traditional caching's inability to dynamically adjust content configuration according to user behavior. It can allocate and adjust hotspot data across different caching layers based on the user's actual access patterns, ensuring that the most frequently accessed content is always stored in the edge processing layer closest to the user, while less frequently accessed content is stored in the cloud server caching layer, thus significantly improving cache hit rate. Simultaneously, through continuous miss rate monitoring and reinforcement learning feedback adjustments, it possesses self-learning and adaptive capabilities, dynamically optimizing caching strategies as user behavior changes, achieving the technical effects of reducing system response latency, improving user access experience, and optimizing resource utilization efficiency.

[0042] Furthermore, by monitoring and acquiring user behavior datasets within historical time windows, user access characteristics are identified and determined, including:

[0043] S21. Monitor and obtain user behavior datasets within a historical time window, wherein the user behavior data includes at least the content accessed, browsing time, and access frequency;

[0044] S22. Divide the user behavior dataset according to the access content to obtain several access contents and several similar behavior datasets;

[0045] S23. Based on browsing time and access frequency, calculate content popularity according to the several similar behavior datasets to obtain several content popularity values ​​for the several accessed contents, wherein the content popularity is positively correlated with browsing time and access frequency;

[0046] S24. Arrange the content popularity in descending order of popularity to generate a content popularity sequence, which serves as a user access feature.

[0047] In one optional implementation, firstly, a user behavior dataset within a historical time window is monitored and acquired. Specifically, within a preset historical time window (e.g., the past 7 days, 15 days, or 30 days), user access behavior information is continuously collected using techniques such as logs, user tracking, and access statistics to form a user behavior dataset. This user behavior data includes at least the accessed content, browsing time, and access frequency. Accessed content records the specific pages, products, articles, etc., accessed by the user; browsing time records the duration the user spends on each piece of content; and access frequency records the number of times the user accesses specific content within the time window.

[0048] Subsequently, based on the access content, the user behavior dataset is categorized and organized according to different content types, resulting in several different access contents and several similar behavior datasets for each access content. For example, all access records for product A are grouped into one similar behavior dataset, all access records for article B are grouped into another similar behavior dataset, and so on, to achieve ordered grouping of the user behavior dataset.

[0049] Next, for each dataset of similar behaviors, a weighted calculation method is used to comprehensively consider the contributions of browsing time and access frequency to content popularity. Specifically, browsing time and access frequency weights are set, multiplied by the corresponding browsing time and access frequency respectively, and then summed to obtain the popularity contribution value of a single access; then, the popularity contribution values ​​of all access records for the same content are accumulated and summed to finally obtain the overall content popularity of that accessed content. Since content popularity is positively correlated with both browsing time and access frequency, the longer the user stays and the more times the content is accessed, the higher its calculated popularity value. In this way, based on several datasets of similar behaviors, the content popularity of several accessed contents is calculated separately, resulting in several content popularity values ​​for several accessed contents.

[0050] Subsequently, the content popularity scores of several accessed content items are arranged in descending order of value to form an ordered content popularity sequence, which serves as a quantified user access characteristic. This content popularity directly reflects the user's level of attention and access preferences for different content; content with higher popularity scores ranks higher, providing a basis for subsequent caching strategy formulation and data updates.

[0051] Furthermore, updating the second hot data based on the user access characteristics includes:

[0052] S25. Obtain a set of historical content popularity sequences within a historical time range, perform content popularity fluctuation analysis based on the set of historical content popularity sequences, obtain a popularity fluctuation coefficient, and correct the initial gradient weight based on the popularity fluctuation coefficient to obtain an adaptive gradient weight.

[0053] S26. Based on the adaptive gradient weight, perform a weighted calculation of popularity according to the historical content popularity sequence set and the content popularity sequence, and obtain a comprehensive content popularity sequence after sorting.

[0054] S27. Select the accessed content corresponding to the top preset proportion of the comprehensive content popularity in the comprehensive content popularity sequence as high-frequency accessed content, and call the complete content data corresponding to multiple high-frequency accessed content in the data center persistence layer as the updated second hot data.

[0055] In a preferred embodiment, firstly, a set of historical content popularity sequences within a historical time range is acquired, and content popularity fluctuation analysis is performed based on this set. Specifically, within a relatively small historical time range (such as the past 3 days, a week, etc.), content popularity sequences generated at various time points within that historical time range are collected to form a set of historical content popularity sequences. By comparing and analyzing the historical content popularity sequences in the set, the degree of fluctuation in content popularity is evaluated, and a popularity fluctuation coefficient is obtained. This popularity fluctuation coefficient reflects the stability of user access behavior; the greater the fluctuation, the more frequently the user access pattern changes, requiring corresponding adjustments to the weight allocation. Based on the popularity fluctuation coefficient, the preset initial gradual weights are corrected and optimized to obtain adaptive gradual weights that can adapt to the characteristics of current user behavior changes.

[0056] Subsequently, a weighted calculation of popularity is performed based on the adaptive gradient weights to generate a comprehensive content popularity sequence. Specifically, the obtained adaptive gradient weights are applied to both the historical content popularity sequence set and the current content popularity sequence, and a weighted fusion calculation is performed on the historical and current data. Because the adaptive gradient weights take into account the fluctuation characteristics of content popularity, they can highlight the importance of current user access characteristics while maintaining the reference value of historical trends. The weighted calculation results are then reordered according to popularity values ​​to obtain a more accurate and comprehensive comprehensive content popularity sequence, which reflects users' historical access preferences and current access needs.

[0057] Next, based on the comprehensive content popularity sequence, frequently accessed content is determined, and the second-hot data is updated. Specifically, a pre-defined percentage (e.g., top 20%, top 30%) of high-popularity content is selected from the comprehensive content popularity sequence. This content is identified as frequently accessed content, representing the hot data that users are most likely to access. Then, based on the identifiers of these frequently accessed content, the corresponding complete content data is retrieved from the data center's persistent layer, including detailed page information, product attributes, and the full text of the article. This retrieved complete content data is used as the updated second-hot data, replacing the original cached content in the cloud server cache layer. This ensures that the data stored in the cloud server cache layer can maximally meet the actual access needs of users, improving cache hit rate and response efficiency.

[0058] Furthermore, based on the historical content popularity sequence set, content popularity fluctuation analysis is performed to obtain a popularity fluctuation coefficient, and the initial gradient weight is corrected based on the popularity fluctuation coefficient, including:

[0059] S251. According to the chronological order, perform similarity comparison on each pair of adjacent historical content popularity sequences in the historical content popularity sequence set, and output the content popularity similarity coefficient sequence.

[0060] S252. Perform content popularity fluctuation analysis on the content popularity similarity coefficient sequence to obtain the popularity fluctuation coefficient, wherein the popularity fluctuation coefficient is the ratio of the standard deviation to the mean of the content popularity similarity coefficients in the content popularity similarity coefficient sequence.

[0061] S253. Configure the initial gradient weight, wherein the weight is negatively correlated with the time interval from the current time, and the initial weight of the current content popularity sequence is 0.3;

[0062] S254. The ratio of the heat fluctuation coefficient to the average heat fluctuation coefficient in the historical time zone is set as the weight adjustment coefficient, and multiplied by the product of the initial weight, is used as the current adaptation weight of the current content heat sequence, wherein the adaptation weight is greater than or equal to 0.3 and less than or equal to 0.7.

[0063] S255. Subtract the current adaptation weight from 1 to obtain the sum of historical adaptation weights;

[0064] S256. Based on the historical adaptation weights and historical popularity weights, a historical adaptation popularity weight sequence is obtained, and the current adaptation weights are combined to generate adaptation gradient weights.

[0065] In a preferred embodiment, firstly, a similarity analysis is performed on the historical content popularity sequences in chronological order. Specifically, the historical content popularity sequences in the set are sorted according to their generation time, and then a similarity comparison is calculated between two temporally adjacent historical content popularity sequences. By calculating the degree of similarity between adjacent historical content popularity sequences in terms of content order, popularity distribution, etc., a numerical index reflecting the similarity between sequences is obtained. The similarity calculation results of all adjacent historical content popularity sequence pairs are arranged sequentially to form a content popularity similarity coefficient sequence. This content popularity similarity coefficient sequence can reflect the continuity and change characteristics of user access behavior in the time dimension.

[0066] Subsequently, a popularity fluctuation coefficient is calculated based on the content popularity similarity coefficient sequence to quantify the stability of user access behavior. Specifically, statistical analysis is performed on all similarity coefficient values ​​in the content popularity similarity coefficient sequence, calculating their standard deviation and mean. The ratio of the standard deviation to the mean is then used as the popularity fluctuation coefficient. This fluctuation coefficient reflects the degree of fluctuation in user access patterns: a larger ratio indicates a higher degree of dispersion in the similarity coefficient, and more unstable changes in user access behavior; a smaller ratio indicates relatively stable user access behavior with strong regularity.

[0067] Next, initial gradient weights are configured, taking into account the time decay characteristics. Specifically, based on the principle that data closer to the current time is more important, the weights are set to be negatively correlated with the time interval from the current time, i.e., the longer the time interval, the smaller the weight. The initial weight of the current content popularity sequence is set to 0.3, while the weight of the historical content popularity sequence decreases with increasing time distance, forming the initial gradient weights. Subsequently, the initial weights are dynamically adjusted based on the popularity fluctuation coefficient. Specifically, the current popularity fluctuation coefficient is compared with the average popularity fluctuation coefficient over a historical time zone (such as the past 3 months, half a year, or other longer periods), and the ratio of the two is calculated as the weight adjustment coefficient. This weight adjustment coefficient reflects the degree of deviation of the current fluctuation from the historical average level. The weight adjustment coefficient is multiplied by the initial weight of the current content popularity sequence to obtain the current adapted weight after fluctuation characteristic correction, and this weight is limited to the range of 0.3 to 0.7 to ensure the rationality of weight allocation.

[0068] Subsequently, the total weights allocated to all historical content popularity sequences are obtained by subtracting the current adaptation weight from 1; this is the historical adaptation weight sum. For example, if the historical time range contains 5 historical content popularity sequences, the historical adaptation weight sum is allocated in chronological order, with historical sequences further back in time receiving smaller weights and those closer to the current time receiving larger weights, forming a historical adaptation popularity weight sequence that conforms to the time decay law. Then, the current adaptation weight is combined with the historical adaptation popularity weight sequence to generate a complete adaptation gradient weight. This adaptation gradient weight considers both the time decay characteristics of user access behavior and dynamically adjusts according to popularity fluctuations, more accurately reflecting the importance of data in different time periods to the current caching strategy, and providing a weight basis for subsequent popularity weighted calculations.

[0069] Furthermore, based on the user access characteristics, an adaptation and dimensionality reduction strategy is set to reduce the dimensionality of the updated second hot data, including:

[0070] S31. Calculate the standard deviation of the content popularity sequence as the popularity distribution deviation amplitude;

[0071] S32. If the deviation of the heat distribution is less than or equal to the preset deviation index, the adaptation dimensionality reduction strategy is set to a proportional dimensionality reduction scheme, wherein the fixed dimensionality reduction ratio is 60%.

[0072] S33. If the deviation of the popularity distribution is greater than the preset deviation index, the adaptation dimensionality reduction strategy is set as an adaptive dimensionality reduction scheme. The adaptive dimensionality reduction scheme includes multiple adaptation dimensionality reduction ratios for multiple frequently accessed content. The adaptation dimensionality reduction ratio is the product of the ratio of the average content popularity of the content popularity sequence to the content popularity of the frequently accessed content and the fixed dimensionality reduction ratio. The adaptation dimensionality reduction ratio is greater than or equal to 40% and less than or equal to 80%.

[0073] In a preferred embodiment, firstly, the distribution characteristics of the content popularity sequence are calculated to determine the magnitude of the popularity distribution deviation. Specifically, statistical analysis is performed on the popularity of all content in the content popularity sequence, and its standard deviation is calculated as the magnitude of the popularity distribution deviation. This magnitude of the popularity distribution deviation reflects the degree of concentration of user access behavior: the smaller the standard deviation, the more similar the popularity of each accessed content, and the more evenly distributed the user access; the larger the standard deviation, the more significant the differences in content popularity, and the more significant the concentration of hot spots in user access.

[0074] Subsequently, the appropriate dimensionality reduction strategy is selected based on the magnitude of the deviation in the popularity distribution. Specifically, the calculated magnitude of the popularity distribution deviation is compared with a preset deviation index. When the magnitude of the popularity distribution deviation is less than or equal to the preset deviation index, it indicates that user access is relatively uniform, and the importance of each accessed content is not significantly different. In this case, the appropriate dimensionality reduction strategy is set to a proportional dimensionality reduction scheme. Under this proportional dimensionality reduction scheme, all frequently accessed content is compressed using a uniform fixed 60% dimensionality reduction ratio. This means that 60% of the key information of the original data is retained, and 40% of redundant content is removed, achieving globally uniform dimensionality reduction processing.

[0075] When the deviation in popularity distribution exceeds a preset deviation index, a differentiated adaptive dimensionality reduction scheme is adopted. Specifically, at this point, there is a clear concentration of hotspots in user access, and the importance of different accessed content varies significantly. Therefore, an adaptive dimensionality reduction scheme is set as the adaptation strategy. In this adaptive dimensionality reduction scheme, the corresponding adaptation dimensionality reduction ratio is calculated for each accessed content based on its specific content popularity. The calculation method is as follows: the ratio of the average content popularity of the content popularity sequence to the content popularity of the specific high-frequency accessed content is calculated, and then this ratio is multiplied by a fixed dimensionality reduction ratio of 60% to obtain the adaptation dimensionality reduction ratio for that content. Since the higher the content popularity, the smaller the ratio, the smaller the final dimensionality reduction ratio, meaning that higher-popularity content retains more original information. At the same time, it is ensured that the adaptive adaptation dimensionality reduction ratio is within a reasonable range. Specifically, the adaptation dimensionality reduction ratio calculated for each high-frequency accessed content is limited to the range of 40% to 80%, ensuring that neither excessive dimensionality reduction results in the loss of key information, nor insufficient dimensionality reduction affects the storage and response efficiency of the edge processing layer. By using adaptive dimensionality reduction, more detailed information can be retained for important content with higher popularity to meet users' precise access needs, while content with relatively lower popularity is streamlined to save storage space and transmission bandwidth, thus achieving personalized data dimensionality reduction optimization.

[0076] Furthermore, if the first data miss rate and / or the second data miss rate are greater than a preset threshold, a cache content adjustment mechanism is triggered, and the updated second hot data and / or the updated first hot data are optimized and adjusted based on reinforcement learning, including:

[0077] S41. Construct a first cache policy optimizer and a second cache policy optimizer based on reinforcement learning;

[0078] S42. If the first data miss rate is greater than a preset threshold and the second data miss rate is less than or equal to a preset threshold, the first cache strategy optimizer is used to perform cache strategy optimization analysis based on the first data miss rate and the adaptation dimensionality reduction strategy, and outputs an optimized dimensionality reduction strategy to optimize and adjust the updated first hot data.

[0079] In a preferred embodiment, firstly, a cache policy optimizer is constructed based on reinforcement learning technology. Specifically, leveraging the autonomous learning and decision-making optimization capabilities of reinforcement learning algorithms, a first cache policy optimizer and a second cache policy optimizer are constructed for different cache layers. The first cache policy optimizer is specifically responsible for optimizing the cache policy of the edge processing layer. By learning the correlation between the first data miss rate and the adaptation dimensionality reduction strategy, it dynamically adjusts the dimensionality reduction parameters to improve the caching effect of the edge layer. The second cache policy optimizer is specifically responsible for optimizing the policy of the cloud server cache layer. By analyzing the interaction between the second data miss rate and the adaptation gradual weight, it adjusts the weight allocation strategy to improve the hit performance of the cloud cache layer. These two cache policy optimizers employ independent learning models, enabling specialized policy optimization tailored to the characteristics of their respective layers.

[0080] Subsequently, specific optimizations were performed to address the poor caching performance of the edge processing layer. Specifically, when monitoring revealed that the first data miss rate exceeded a preset threshold, while the second data miss rate remained within the normal range (less than or equal to the preset threshold), it indicated that the problem primarily occurred at the edge processing layer. This meant the current dimensionality reduction strategy might be inadequate, causing the first hot data to fail to effectively meet user access needs. At this point, the first cache strategy optimizer was activated, using the first data miss rate as feedback and conducting in-depth analysis in conjunction with the currently used adaptive dimensionality reduction strategy. The first cache strategy optimizer used a reinforcement learning algorithm to evaluate the impact of different dimensionality reduction parameters on the cache hit rate, learning the optimal dimensionality reduction ratio, content selection criteria, and other key parameters, and outputting an intelligently optimized dimensionality reduction strategy, thus obtaining the optimized dimensionality reduction strategy.

[0081] Next, the first hot data is readjusted based on the optimized dimensionality reduction strategy. Specifically, the optimized dimensionality reduction strategy output by the first cache strategy optimizer is applied to the dimensionality reduction process of updating the second hot data, redetermining key parameters such as the dimensionality reduction ratio of each frequently accessed content and the data simplification scheme. Through more precise dimensionality reduction, first hot data that better meets the actual access needs of users is generated, and the optimized data is redeployed to the edge processing layer. This dynamic optimization mechanism based on reinforcement learning can continuously improve strategy parameters according to actual cache performance, achieving adaptive cache optimization and performance improvement.

[0082] Furthermore, embodiments of this application also include:

[0083] S43. If the second data miss rate is greater than a preset threshold, the second cache strategy optimizer is used to perform cache strategy optimization analysis based on the second data miss rate and the adaptive gradient weight, output the optimized gradient weight, and optimize and adjust the updated second hot data to obtain the optimized second hot data.

[0084] S44. Based on the optimized second hot data, the updated first hot data is optimized and adjusted according to the optimization dimensionality reduction strategy.

[0085] In a preferred embodiment, specific optimization is performed to address the issue of poor caching performance in the cloud server caching layer. Specifically, when monitoring reveals that the second data miss rate exceeds a preset threshold, it indicates that the second most popular data stored in the cloud server caching layer can no longer effectively meet user access needs, requiring adjustment of the cache content selection strategy. At this point, the second cache strategy optimizer is activated, using the second data miss rate as a performance feedback indicator and combining it with the currently used adaptive gradient weights for in-depth cache strategy optimization analysis. The second cache strategy optimizer uses reinforcement learning algorithms to analyze the impact of different weight configurations on the cache hit rate, learns the optimal weight allocation scheme for historical and current data, and the importance evaluation criteria for data in each time period, ultimately outputting an intelligently optimized gradient weight configuration scheme, resulting in optimized gradient weights. Based on these optimized gradient weights, the popularity weighting calculation and content selection are recalculated, and the updated second most popular data is optimized and adjusted to obtain optimized second most popular data that better matches the user's current access pattern.

[0086] Subsequently, the edge processing layer is adjusted in conjunction with the optimized second hot data. Specifically, since the first hot data of the edge processing layer is obtained by dimensionality reduction of the second hot data of the cloud server cache layer, the first hot data needs to be regenerated accordingly after the second hot data is optimized. The optimized second hot data is used as a new data source, and combined with the optimization dimensionality reduction strategy output by the first cache strategy optimizer in the previous steps, the first hot data is re-dimensionalized and optimized. This layered linkage mechanism, which first adjusts the cloud server cache layer and then optimizes the edge processing layer based on the adjustment results, ensures the consistency and synergy of the entire caching system, avoids data mismatch problems that may be caused by independent optimization of each layer, and realizes end-to-end cache optimization from the data source to the edge node, maximizing the overall cache hit rate and response efficiency.

[0087] Example 2, as Figure 2 As shown, based on the same inventive concept as the software server configuration method for user usage feature analysis provided in Embodiment 1, this embodiment of the invention also provides a software server configuration system for user usage feature analysis, comprising:

[0088] The layered architecture construction module 11 is used to construct a layered distributed content service architecture, wherein the layered distributed content service architecture includes an edge processing layer, a cloud server caching layer and a data center persistence layer, the edge processing layer is used to store first hot data and the cloud server caching layer is used to store second hot data.

[0089] User behavior monitoring module 12 is used to monitor and acquire user behavior datasets within a historical time window, identify and determine user access characteristics, update the second hot data according to the user access characteristics, and store the updated second hot data in the cloud server cache layer.

[0090] Data dimensionality reduction processing module 13 is used to set an adaptation dimensionality reduction strategy according to the user access characteristics, reduce the dimensionality of the updated second hot data, and obtain the updated first hot data to be stored in the edge processing layer.

[0091] The cache optimization and adjustment module 14 is used to continuously monitor the cache hit status of user access requests, obtain the first data miss rate and the second data miss rate within a preset time zone, and if the first data miss rate and / or the second data miss rate are greater than a preset threshold, trigger the cache content adjustment mechanism to optimize and adjust the updated second hot data and / or the updated first hot data based on reinforcement learning.

[0092] Furthermore, the execution steps of the user behavior monitoring module 12 include:

[0093] Monitor and acquire user behavior datasets within historical time windows, including at least accessed content, browsing time, and access frequency.

[0094] The user behavior dataset is divided according to the access content to obtain several access contents and several similar behavior datasets;

[0095] Based on browsing time and access frequency, content popularity is calculated according to several similar behavior datasets to obtain several content popularity values ​​for the accessed content. Among them, content popularity is positively correlated with browsing time and access frequency.

[0096] The content popularity is arranged in descending order of popularity to generate a content popularity sequence, which serves as a user access feature.

[0097] Furthermore, the execution steps of the user behavior monitoring module 12 also include:

[0098] Obtain a set of historical content popularity sequences within a historical time range, perform content popularity fluctuation analysis based on the set of historical content popularity sequences, obtain a popularity fluctuation coefficient, and correct the initial gradient weight based on the popularity fluctuation coefficient to obtain an adaptive gradient weight.

[0099] Based on the adaptive gradient weight, a weighted calculation of popularity is performed according to the historical content popularity sequence set and the content popularity sequence, and the comprehensive content popularity sequence is obtained after sorting.

[0100] The accessed content corresponding to the top preset proportion of the comprehensive content popularity in the comprehensive content popularity sequence is selected as the high-frequency accessed content. The complete content data corresponding to multiple high-frequency accessed content in the persistent layer of the data center is called as the updated second hot data.

[0101] Furthermore, the execution steps of the user behavior monitoring module 12 also include:

[0102] Based on the chronological order, similarity comparisons are performed on each pair of adjacent historical content popularity sequences in the historical content popularity sequence set, and a content popularity similarity coefficient sequence is output.

[0103] Content popularity fluctuation analysis is performed on the content popularity similarity coefficient sequence to obtain the popularity fluctuation coefficient, wherein the popularity fluctuation coefficient is the ratio of the standard deviation to the mean of the content popularity similarity coefficients in the content popularity similarity coefficient sequence;

[0104] Configure initial gradient weights, where the weights are negatively correlated with the time interval from the current time, and the initial weight of the current content popularity sequence is 0.3;

[0105] The ratio of the heat fluctuation coefficient to the average heat fluctuation coefficient in the historical time zone is set as the weight adjustment coefficient, and multiplied by the initial weight as the current adaptation weight of the current content heat sequence, wherein the adaptation weight is greater than or equal to 0.3 and less than or equal to 0.7.

[0106] Subtracting the current adaptation weight from 1 yields the sum of historical adaptation weights;

[0107] Based on the historical adaptation weights and the historical popularity weights, a historical adaptation popularity weight sequence is obtained, and an adaptation gradient weight is generated by combining the current adaptation weights.

[0108] Furthermore, the execution steps of the data dimensionality reduction processing module 13 include:

[0109] Calculate the standard deviation of the content popularity sequence as the magnitude of the popularity distribution deviation;

[0110] If the deviation of the heat distribution is less than or equal to the preset deviation index, the adaptation dimensionality reduction strategy is set to a proportional dimensionality reduction scheme, wherein the fixed dimensionality reduction ratio is 60%.

[0111] If the deviation of the popularity distribution is greater than the preset deviation index, the adaptation dimensionality reduction strategy is set as an adaptive dimensionality reduction scheme. The adaptive dimensionality reduction scheme includes multiple adaptation dimensionality reduction ratios for multiple frequently accessed content. The adaptation dimensionality reduction ratio is the product of the ratio of the average content popularity of the content popularity sequence to the content popularity of the frequently accessed content and the fixed dimensionality reduction ratio, and the adaptation dimensionality reduction ratio is greater than or equal to 40% and less than or equal to 80%.

[0112] Furthermore, the execution steps of the cache optimization and tuning module 14 include:

[0113] A first cache policy optimizer and a second cache policy optimizer are constructed based on reinforcement learning;

[0114] If the first data miss rate is greater than a preset threshold and the second data miss rate is less than or equal to a preset threshold, the first cache strategy optimizer is used to perform cache strategy optimization analysis based on the first data miss rate and the adaptation dimensionality reduction strategy, and outputs an optimized dimensionality reduction strategy to optimize and adjust the updated first hot data.

[0115] Furthermore, the execution steps of the cache optimization and tuning module 14 also include:

[0116] If the second data miss rate is greater than a preset threshold, the second cache strategy optimizer is used to perform cache strategy optimization analysis based on the second data miss rate and the adaptive gradient weight, output the optimized gradient weight, and optimize and adjust the updated second hot data to obtain the optimized second hot data.

[0117] Based on the optimized second hot data, the updated first hot data is optimized and adjusted according to the optimization dimensionality reduction strategy.

[0118] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0119] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0120] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0121] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0122] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0123] Although preferred embodiments of the invention have been described, those skilled in the art, once they have learned the basic inventive concept, can make other changes and modifications to these embodiments.

[0124] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A software server configuration method for user characteristic analysis, characterized in that, The method includes: A layered distributed content service architecture is constructed, comprising an edge processing layer, a cloud server caching layer, and a data center persistence layer. The edge processing layer is used to store first hot data, and the cloud server caching layer is used to store second hot data. The first hot data is high-frequency access data that has undergone dimensionality reduction processing, and the second hot data is complete data of high-frequency access content identified based on user access characteristics. The system monitors and acquires user behavior datasets within a historical time window, identifies and determines user access characteristics, updates the second hot data based on the user access characteristics, and stores the updated second hot data in the cloud server cache layer. Based on the user access characteristics, an adaptation and dimensionality reduction strategy is set to reduce the dimensionality of the updated second hot data and obtain the updated first hot data to be stored in the edge processing layer. Continuously monitor the cache hit status of user access requests, obtain the first data miss rate and the second data miss rate within a preset time zone, and if the first data miss rate and / or the second data miss rate are greater than a preset threshold, trigger the cache content adjustment mechanism, and optimize and adjust the updated second hot data and / or the updated first hot data based on reinforcement learning. Monitor and acquire user behavior datasets within historical time windows to identify and determine user access characteristics, including: Monitor and acquire user behavior datasets within historical time windows, including at least accessed content, browsing time, and access frequency. The user behavior dataset is divided according to the access content to obtain several access contents and several similar behavior datasets; Based on browsing time and access frequency, content popularity is calculated according to several similar behavior datasets to obtain several content popularity values ​​for the accessed content. Among them, content popularity is positively correlated with browsing time and access frequency. The content popularity is arranged in descending order of popularity to generate a content popularity sequence, which serves as a user access feature. Based on the user access characteristics, an adaptation and dimensionality reduction strategy is set to reduce the dimensionality of the updated second hot data, including: Calculate the standard deviation of the content popularity sequence as the magnitude of the popularity distribution deviation; If the deviation of the heat distribution is less than or equal to the preset deviation index, the adaptation dimensionality reduction strategy is set to a proportional dimensionality reduction scheme, wherein the fixed dimensionality reduction ratio is 60%; If the deviation of the popularity distribution is greater than the preset deviation index, the adaptation dimensionality reduction strategy is set as an adaptive dimensionality reduction scheme. The adaptive dimensionality reduction scheme includes multiple adaptation dimensionality reduction ratios for multiple frequently accessed content. The adaptation dimensionality reduction ratio is the product of the ratio of the average content popularity of the content popularity sequence to the content popularity of the frequently accessed content and the fixed dimensionality reduction ratio, and the adaptation dimensionality reduction ratio is greater than or equal to 40% and less than or equal to 80%.

2. The software server configuration method for user usage feature analysis according to claim 1, characterized in that, Updating the second hot data based on the user access characteristics includes: Obtain a set of historical content popularity sequences within a historical time range, perform content popularity fluctuation analysis based on the set of historical content popularity sequences, obtain a popularity fluctuation coefficient, and correct the initial gradient weight based on the popularity fluctuation coefficient to obtain an adaptive gradient weight. Based on the adaptive gradient weight, a weighted calculation of popularity is performed according to the historical content popularity sequence set and the content popularity sequence, and the comprehensive content popularity sequence is obtained after sorting. The accessed content corresponding to the top preset proportion of the comprehensive content popularity in the comprehensive content popularity sequence is selected as the high-frequency accessed content. The complete content data corresponding to multiple high-frequency accessed content in the persistent layer of the data center is called as the updated second hot data.

3. The software server configuration method for user usage feature analysis according to claim 2, characterized in that, Based on the historical content popularity sequence set, content popularity fluctuation analysis is performed to obtain a popularity fluctuation coefficient, and the initial gradient weight is corrected based on the popularity fluctuation coefficient, including: Based on the chronological order, similarity comparisons are performed on each pair of adjacent historical content popularity sequences in the historical content popularity sequence set, and a content popularity similarity coefficient sequence is output. Content popularity fluctuation analysis is performed on the content popularity similarity coefficient sequence to obtain the popularity fluctuation coefficient, wherein the popularity fluctuation coefficient is the ratio of the standard deviation to the mean of the content popularity similarity coefficients in the content popularity similarity coefficient sequence; Configure initial gradient weights, where the weights are negatively correlated with the time interval from the current time, and the initial weight of the current content popularity sequence is 0.3; The ratio of the heat fluctuation coefficient to the average heat fluctuation coefficient in the historical time zone is set as the weight adjustment coefficient, and multiplied by the initial weight as the current adaptation weight of the current content heat sequence, wherein the adaptation weight is greater than or equal to 0.3 and less than or equal to 0.

7. Subtracting the current adaptation weight from 1 yields the sum of historical adaptation weights; Based on the historical adaptation weights and the historical popularity weights, a historical adaptation popularity weight sequence is obtained, and an adaptation gradient weight is generated by combining the current adaptation weights.

4. The software server configuration method for user feature analysis according to claim 1, characterized in that, If the miss rate of the first data and / or the miss rate of the second data are greater than a preset threshold, a cache content adjustment mechanism is triggered. Based on reinforcement learning, the updated second hot data and / or the updated first hot data are optimized and adjusted, including: A first cache policy optimizer and a second cache policy optimizer are constructed based on reinforcement learning; If the first data miss rate is greater than a preset threshold and the second data miss rate is less than or equal to a preset threshold, the first cache strategy optimizer is used to perform cache strategy optimization analysis based on the first data miss rate and the adaptation dimensionality reduction strategy, and outputs an optimized dimensionality reduction strategy to optimize and adjust the updated first hot data.

5. The software server configuration method for user feature analysis according to claim 4, characterized in that, If the second data miss rate is greater than a preset threshold, the second cache strategy optimizer is used to perform cache strategy optimization analysis based on the second data miss rate and the adaptive gradient weight, output the optimized gradient weight, and optimize and adjust the updated second hot data to obtain the optimized second hot data. Based on the optimized second hot data, the updated first hot data is optimized and adjusted according to the optimization dimensionality reduction strategy.

6. A software server configuration system for user feature analysis, characterized in that, A software server configuration method for implementing a user usage feature analysis method as described in any one of claims 1 to 5, the system comprising: A layered architecture building module is used to build a layered distributed content service architecture, wherein the layered distributed content service architecture includes an edge processing layer, a cloud server caching layer, and a data center persistence layer. The edge processing layer is used to store first hot data, and the cloud server caching layer is used to store second hot data. The user behavior monitoring module is used to monitor and acquire user behavior datasets within a historical time window, identify and determine user access characteristics, update the second hot data according to the user access characteristics, and store the updated second hot data in the cloud server cache layer. The data dimensionality reduction processing module is used to set an adaptation dimensionality reduction strategy according to the user access characteristics, reduce the dimensionality of the updated second hot data, and obtain the updated first hot data to be stored in the edge processing layer. The cache optimization and adjustment module is used to continuously monitor the cache hit status of user access requests, obtain the first data miss rate and the second data miss rate within a preset time zone, and if the first data miss rate and / or the second data miss rate are greater than a preset threshold, trigger the cache content adjustment mechanism to optimize and adjust the updated second hot data and / or the updated first hot data based on reinforcement learning.

Images